Enter a new value for the quasi-likelihood parameter (c-hat).  Typically, this value is 1.0000, meaning that the data are not over-dispersed, i.e., no extrabinomial variation.  However, you may want to correct for overdispersion by increasing the value.  An estimator of c-hat is the deviance divided by its degrees of freedom, although this estimator generally is biased high for finite sample sizes.

The value for c cannot be <1, because there is no biologically reasonable model that would generate underdispersed data.

The value of c-hat is used to compute the QAIC (quasi-AIC) for a model with k parameters by the following formula:

QAIC = -2log Likelihood/c-hat + 2K

and the QAIC by the formula:

QAICc = -2log Likelihood/c-hat + 2K + 2K(K + 1)/(n-ess – K – 1)

where n-ess is the effective sample size.

Values of c-hat > 1 can also be used to compute profile likelihod confidense intervals.

Methods in MARK to estimate c are the median chat and the Fletcher chat estimators.

More details on c-hat are provided on the WWW page